- Published on
Morgan Stanley cut its riskiest reconciliation job in half — by making its agents less autonomous
- Authors
- Name
- DeepHorus
Morgan Stanley has significantly reduced the time required for profit and loss (P&L) reconciliation by deploying an agentic system that prioritizes human oversight over full autonomy. The system, known as FIXR, has cut the riskiest reconciliation job in half by making its agents less autonomous, a counterintuitive approach that has proven highly effective. This deployment targets one of banking's most accuracy-critical and deadline-driven workflows, where even minor errors can have substantial financial implications. The move underscores a growing trend in enterprise AI: using tightly controlled automation to enhance, rather than replace, human expertise.
Every trading day, Morgan Stanley's trade desks handle transactions such as cash equities and debt investments, requiring controllers to reconcile P&L across multiple systems. With hundreds of thousands of attributes frequently failing to match, controllers must manually investigate each discrepancy, or 'break,' making decisions and signing off before a hard morning deadline. Previously, this process could take up to six hours for a single book, but FIXR now completes the task in two to three hours. Across roughly 100 controllers, this translates to approximately 1,500 hours saved per week.
FIXR operates through multiple agents that work collaboratively: one interprets past guidance to propose resolutions, another learns from controller behavior and documents rules, and a third converts repeated patterns into automated logic. The system can auto-clear familiar breaks, suggest solutions for novel ones, and flag items for human investigation. Critically, humans remain fully in the loop, reviewing, approving, or correcting every recommendation, and feeding decisions back to improve future runs. 'You still preserve that element of human accountability even as you start to automate,' said Morgan Stanley Managing Director Todd Johnson at a recent VB AI Impact event.
The decision to limit autonomy was deliberate, as autonomy requires a high degree of trust that may not be warranted in high-stakes environments. Johnson emphasized that enterprises will not see efficiency gains if everyone is checking everything an agent does, but the human–agent feedback loop was essential for controlled, measured, and repeatable automation. 'We recognized that all that intelligence that's sitting in the mind of a controller is gonna be difficult to get all into an agent on day one,' he noted. This approach ensures that the system learns incrementally from human expertise, building trust over time.
Johnson highlighted the importance of establishing processes before integrating AI. His team conducted a thorough process intelligence assessment to map workflows and identify where automation would be most advantageous. 'If we can fix that first before we add agents to the problem, then we really will be transforming the opportunity,' he said. The P&L sign-off process was full of manual steps suitable for automation, and agents taking over these tasks free up controllers for more value-added work.
Looking ahead, Morgan Stanley plans to expand FIXR to other complex workflows, leveraging the same human-in-the-loop framework. The system's ability to codify controller knowledge into repeatable rules suggests a future where AI agents handle routine decisions while humans focus on exceptions and strategic analysis. As Johnson put it, 'Over time you'll see more and more of those items resolved in an automatic way.' This balanced approach may become a model for enterprises seeking to deploy AI in mission-critical domains without sacrificing accuracy or accountability.
